Add optimized Poseidon implementation

[mpenciak]

We should implement the optimizations described in the white paper.

These optimizations boil down to two steps:

• Swap the order of the linear and ARC layers, and pre-calculate the action of the MDS matrix on the round constants.
• Factorize the MDS matrix into a pre-sparse matrix and a sparse matrix. Their action on the state vector can be calculated in an efficient manner using sparse matrix algorithms.

(see appendix B of the white paper: https://eprint.iacr.org/2019/458.pdf)

This should also include a robust test suite to ensure the optimized hash implementation matches with the naiive "unoptimized" implementation.

See:

• Python implementation: https://github.com/ingonyama-zk/poseidon-hash/blob/1b72e87238e95d7ed8be63bd7c231b1d3a7936f3/poseidon/hash.py#L144
• Optimized reference implementation: https://extgit.iaik.tugraz.at/krypto/hadeshash/-/blob/master/code/poseidonperm_x3_64_24_optimized.sage
• Rust implementation: https://github.com/filecoin-project/neptune/blob/2b11f0ce69f52aa9594f250baa658bfe2d349ac3/src/poseidon.rs#L389
• The filecoin specification: https://spec.filecoin.io/algorithms/crypto/poseidon/#section-algorithms.crypto.poseidon.optimized-poseidon